speculative

Coordinate parallel implementations and select the best solution via weighted criteria.

1|Updated Jan 13, 2026

Search Tags:#evaluation #agents #orchestration #parallel #judge #speculative

Ready to run

npx skills add https://github.com/wgordon17/personal-claude-marketplace --skill speculative

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: speculative
Source: https://github.com/wgordon17/personal-claude-marketplace/tree/main/code-quality/skills/speculative
Command: npx skills add https://github.com/wgordon17/personal-claude-marketplace --skill speculative

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Coordinating multiple viable approaches to complex problems by running parallel implementations, evaluating them with a formal judge, and selecting the best solution.

Core Features & Use Cases

Orchestrates 2-4 competing implementations in isolated workspaces with a unified evaluation framework.
Uses a weighted success-criteria model to rank results and determine the winner.
Produces auditable artifacts: implementation results, judge judgments, and a final speculative report.

Quick Start

Define the task and success criteria, spawn competing agents in isolated worktrees, and let the judge pick the winner.

Frequently Asked Questions about speculative

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I run parallel implementations to choose the best solution for a problem?▼

Parallel implementations are orchestrated by spawning 2-4 competing agents in isolated workspaces, evaluating their results with a weighted success-criteria model, and selecting the highest-ranked solution.

What is the best way to evaluate competing agents and avoid guesswork?▼

Evaluating competing agents without guesswork requires a formal judge that applies weighted success criteria to rank the results, ensuring the winning solution is selected based on formal evaluation rather than intuition.

How do I orchestrate multiple agents in isolated workspaces for evaluation?▼

Orchestrating multiple agents involves spawning 2-4 competing implementations in isolated worktrees, executing them in parallel, and capturing their output as auditable artifacts for a unified evaluation framework.

Can I define custom weighted success criteria for parallel agent orchestration?▼

Yes, you can define custom weighted success criteria by creating a SpeculativeSpec that establishes the evaluation framework used by the judge to rank competing implementations and determine the winner.

What artifacts are produced when running parallel competing implementations?▼

Running parallel competing implementations produces auditable artifacts including the individual implementation results, the judge's formal judgments based on weighted criteria, and a final speculative report summarizing the evaluation.

When should I use a speculative approach instead of a single agent implementation?▼

Use a speculative approach when multiple viable options exist for a complex problem and formal evaluation is needed to avoid guesswork, rather than relying on a single implementation path.